Different pattern: a new EEG-based method for mental performance detection

Area of Science:

• Neuroscience
• Artificial Intelligence
• Signal Processing

Background:

• Electroencephalography (EEG) signals are crucial for understanding brain activity.
• Detecting mental performance from EEG data requires advanced feature extraction and classification methods.
• Existing methods may lack interpretability, limiting their application in neuroscience.

Purpose of the Study:

• To develop a novel feature extraction method, Different Pattern (DiffPat), for enhanced EEG mental performance analysis.
• To integrate DiffPat within an Explainable Feature Engineering (XFE) framework for both classification and explanation.
• To improve the accuracy and interpretability of mental performance detection from EEG signals.

Main Methods:

• Utilized an EEG mental performance dataset.
• Developed the DiffPat algorithm for efficient EEG channel difference-based feature extraction.
• Employed iterative Neighborhood Component Analysis (INCA) for feature selection.
• Applied K-Nearest Neighbor (KNN) classifier for performance classification.
• Generated AI-based explainable results using XAI and Directed Lobish (DLob) for cortical connectome diagrams and interpretive sentences.

Main Results:

• Achieved 84.48% accuracy using leave-one-subject-out (LOSO) cross-validation for subject-independent performance estimation.
• Attained 99.87% accuracy with subject-aware 10-fold cross-validation.
• Demonstrated the model's capability for both accurate classification and result explanation.

Conclusions:

• The DiffPat-based XFE model offers significant advancements in feature engineering and neuroscience.
• The model's dual capability of classification and explanation holds strong potential for mental performance analysis.
• This approach is applicable to a wide range of EEG-based applications beyond mental performance detection.